Apart from the well-known advantages of conventional laparoscopic surgery (LAP) and robotic-assisted surgery (RAS) for the patient, these techniques have a steep learning curve and are physically and mentally demanding for surgeons, as well as presenting ergonomic challenges. Consequently, this mental and physical burden, combined with potential ergonomic deficiencies in the working environment, can lead to musculoskeletal problems in surgeons [1]. Although ergonomic conditions are considered to have improved in RAS compared to LAP, mainly for the primary surgeon operating from the console, scientific evidence remains scarce [2]. Therefore, further ergonomic analysis studies are needed to thoroughly investigate the ergonomic conditions of the surgeon during RAS, identify potential deficiencies, and develop possible solutions and recommendations.

Minimally invasive surgery (MIS) and, in particular, RAS have grown rapidly in recent decades, and several robotic-assisted procedures have become standard surgical techniques [3]. The advantages of MIS and RAS for patients are widely known, including reduced tissue trauma, better cosmetic outcomes, and reduced hospital stay [4]. However, these techniques present some limitations for the surgeon, such as possible ergonomic deficiencies during long surgeries, or high levels of stress in certain procedures [5], which can be detrimental to the surgeon’s health and have an impact on the quality of surgical procedures and patient care [6].

Surgical robotics offers greater surgical precision, better visual perception by allowing 3D images to be obtained compared to the 2D images of LAP, and better ergonomics for the surgeon, who can operate in a seated position. There are some traditional methods to assess the workload, both mental and physical, of surgeons during their surgical activity, mainly oriented to subjective evaluations such as, for example, the SURG-TLX scales [7]. However, the evolution and miniaturization of sensors have allowed the increasing incorporation of wearable technology in ergonomic and physiological analysis in the surgical setting, which has made it possible to offer objective solutions without interrupting the surgeons’ surgical practice. Some of these objective parameters are the surgeon’s posture through body movement techniques, force/torque analysis [8], muscle activity through electromyography (EMG) signal analysis [9], and stress level through electrocardiogram examination or electrodermal activity (EDA) signals [10].

The application of artificial intelligence (AI) has grown exponentially in its use and development. These algorithms are based on nontrivial processes to discover potentially useful knowledge initially hidden in the data [11,12]. Among the different AI techniques, there are several algorithms that allow the development of predictive models [13,14]. Thus, the use of these predictive models in the field of ergonomics during surgical practice offers us a wide range of possibilities to predict those risk situations related to the surgeon’s health, for example, inadequate posture, the appearance of muscle fatigue, or high levels of stress. Being able to predict and, therefore, reduce these risk situations would potentially improve the health of surgeons and, consequently, the quality of surgical practice.

Several studies have been presented focusing on the ergonomic analysis in RAS. Some of these studies use subjective surveys [15], demonstrating that 56.1% of regularly practicing robotic surgeons continue to experience related physical symptoms or discomfort, including neck stiffness, finger and eye fatigue, among the most common.

Other studies analyzed the surgeon’s posture at the console during surgical practice using traditional photogrammetry methodologies [16]. Ergonomically risky positioning of the neck and elbow was observed in medical students, and robotic experts showed risky positioning in the knee and hip region. Another study objectively assessed ergonomics in surgical robotics using rapid entire body assessment and the rapid upper limb assessment to quantify ergonomic effectiveness [17]. An Xbox Connect camera was used to record the surgeon’s posture at the console. The results indicated a medium risk, with the recommendation that measures need to be taken to improve surgeon ergonomics. However, these photogrammetry-based methods are prone to occlusion problems in such cumbersome environments as operating rooms.

In order to deal with the above limitations, wearable systems such as inertial measurement unit–based systems are used. These systems continuously track the surgeon’s movements without interfering with the sterile environment. In a study during the performance of robotic prostatectomies, it was concluded that the console can limit postures, causing static loads that have been associated with musculoskeletal symptoms for the surgeon’s neck, torso, and shoulders [18].

Other studies have objectively analyzed the surgeon’s muscle loading using surface EMG systems [9]. The results indicated that laparoscopic practice presented more muscle fatigue in the forearm compared to robotic-assisted laparoscopic surgery. Other studies analyzed physiological parameters to examine the surgeon’s health, such as heart rate as a possible indicator of stress [16]. The highest level of stress was obtained in surgical residents. In previous work, we used different wearable technologies to record the posture, muscle activity, EDA, and electrocardiographic signal of surgeons during robotic-assisted surgical practice [19]. The results indicated that robot-assisted procedures showed better ergonomic outcomes for the surgeon compared to LAP.

Regarding the use of predictive models in ergonomics in MIS based on AI techniques, in previous studies, we developed predictive models to predict stress from ergonomic (kinematic, body posture, and angular rate) and physiological (EDA, blood pressure, and body temperature) parameters of the surgeon [5] during RAS procedures. The linear models (multiple linear regression [MLR]) proposed in this study were successfully validated, improving the results achieved by the nonlinear models (support vector machine, multilayer perceptron [MLP]), demonstrating the possibility of predicting factors that help us to improve the surgeon’s health during robotic surgery.

Other studies analyzed and compared the surgeon’s posture between RAS and LAP techniques using motion analysis technology with inertial sensors and developed a predictive model of ergonomically inappropriate postures for the surgeon [20]. MLR and MLP were compared, with the best results obtained when applying MLR as a predictive technique.

In this study, the predictive model developed would allow the evaluation of localized muscle fatigue for RAS using MLR and MLP. To do this, we will use novel wearable technologies in the surgical field in order to study the surgeon’s localized muscle fatigue and to develop predictive models based on AI that enable predicting situations and factors that lead to increased muscle fatigue in the field of MIS, both in LAP surgery and RAS.

This work represents a significant step in the application of cutting-edge technology for the improvement of ergonomic conditions in the surgical field, with the aim of improving the surgeon’s health during surgery, contributing to the improvement of surgical performance and, consequently, the quality of patient care.

The main objective of this study is to develop and validate a predictive model of localized muscle fatigue in surgeons based on EMG sensor data during conventional and robot-assisted laparoscopic surgery, using AI techniques (MLP and MLR) and considering different levels of experience (experienced and novice surgeons). The hypothesis of this study is that it would be possible to predict the localized muscle fatigue in MIS by combining wearable technology and AI techniques, with RAS surgery leading to less muscle fatigue for novice surgeons, who would be less conditioned by the habit of LAP practice.

In addition, this study seeks to take a step forward in the prevention and improvement of ergonomic conditions of surgeons, to seek the improvement of their health and, consequently, the quality of surgery and health care of patients.

In this study, 8 different surgeons with different levels of experience in LAP performed 4 different surgical tasks by LAP and RAS, all of them on a task-simulator training model during 25 surgical sessions.

For each surgical session, the EMG signal was recorded by using a wireless system (EMG TRIGNO Avanti System, Delsys, Natick, Massachusetts, USA) on 7 muscle groups bilaterally: Brachioradialis, Brachii Triceps, Erector Spinae, Gastrocnemius Medial, Middle Trapezius, Upper Trapezius, and Vastus Lateralis.

From these data, a dataset (140,198 records) was generated, on which a preprocessing technique (scaling) has been applied, and then, 2 datasets were generated: 80% of the data for the training dataset and the cross-validation of 10-folds (112,158 records), and the remaining 20% of the data for the test dataset (28,040 records). Records from all 25 sessions appeared in both datasets.

In this study, 2 predictive AI approaches were applied to the training dataset to generate the predictive models: MLP and MLR.

Finally, to validate the generated predictive models, they were validated on the 10-fold cross-validation and test dataset. Figure 1 shows the experimental design followed by this study.

Written informed consent was taken and signed by all participants for the record and use of data generated in this study. In this informed consent, the participants could select to opt out of the study or to participate.

As this research focused only on recording the data that did not contain identifying information and posed no additional risk to subjects and participants, its activities were deemed exempt from ethics committee review. This study was conducted under the guidelines of the Spanish normative 14/2007, of July 3, on Biomedical research [25], since this study does not involve human biological samples or critical data that allow the identification of the participants, and it is not invasive for the human participants.

The wireless EMG TRIGNO Avanti System from DELSYS was used to record the muscle activity by means of EMG signals (Figure 3). This system has up to 16 sensors with a sampling rate of 1024 Hz. The EMG signal of the following 7 muscle groups was recorded bilaterally: Brachioradialis, Brachii Triceps, Erector Spinae, Gastrocnemius Medial, Middle Trapezius, Upper Trapezius, and Vastus Lateralis. The EMG sensors were placed on each muscle according to the SENIAM guidelines [26]. These guidelines are widely used in scientific literature, providing detailed recommendations for sensor placement in more than 30 muscles. Among the guidelines, it can be found longitudinal and transverse location, sensor orientation, and interelectrode distance, among others. Prior to the placement of each sensor, the skin was cleaned by gently rubbing it with 70% isopropyl alcohol.

Raw EMG signals were initially processed using a band-pass filter of 20-300 Hz. The filtered EMG was then smoothed with a moving window of 125 milliseconds and calculated as root-mean-square in order to remove artifacts and noise. To normalize the results for each subject, the EMG values were presented as a percentage of maximum voluntary contraction (%MVC). %MVC was computed separately for each muscle group, just before each test, by asking each participant to perform specific contractions against a fixed resistance. The MVC was performed separately for each muscle group just before each test, asking each subject to perform specific contractions against a fixed resistance.

These contractions are performed with the aim of measuring the maximum effort of each muscle group and normalizing the EMG signals of each subject.

Eight surgeons participated in this study: 4 experts in laparoscopic surgery (> 100 laparoscopic surgeries performed; 1.80, SD 0.08 m tall and 36.5, SD 11.03 years old) and 4 novices in laparoscopic surgery (< 100 laparoscopic surgeries performed; 1.82, SD 0.04 m tall and 28.5, SD 7.59 years old). All participants had no previous experience in robotic-assisted laparoscopic surgery. They performed 8 sessions by means of LAP techniques and 17 sessions using RAS techniques.

The results show the comparison of kinematic data as a function of the surgical technique and the level of expertise of the surgeon, as well as the goodness of the predictive models to predict the localized muscle fatigue for each combination of surgical technique and level of expertise on each muscle group analyzed. These results demonstrate the influence of previous experience in LAP (without previous experience in RAS) on the onset of localized muscle fatigue in robot-assisted LAP practice [38].

The muscles analyzed in this study have been extensively evaluated in previous studies [19,39] for LAP and RAS, with the main muscles analyzed being the upper trapezius, middle trapezius, and triceps brachii.

Considering the type of surgical technique, the expert surgeons had less muscle activity with the RAS technique for the dissection, peg transfer, and suturing tasks, mainly for the forearm muscles (triceps brachii, middle trapezius, and upper trapezius). However, the labyrinth task had less muscle activity with the LAP technique for the nondominant hand. Conversely, the dominant hand showed less muscle activity in the RAS technique. For the novice surgeons, the middle trapezius bilaterally and the right upper trapezius had less muscle activity in LAP for the dissection, labyrinth, and pin transfer tasks. For the other cases, they had less muscle activity using the RAS technique. In all cases, the muscle activity was lower using the RAS technique during the suturing task. The results obtained during this study agree with those obtained in previous studies [19,40], which demonstrated that from an ergonomic point of view, RAS is more comfortable than LAP [41].

In the same way, muscle activity is closely related to muscle fatigue, and prolonged muscle activity of small, single muscle fibers can cause degenerative muscle changes even at very low levels of muscle activity. For this reason, proper ergonomics is very important for LAP activities [42]. This fact reinforces the need for future surgeons in surgical robotics to receive structured, progressive, and adequate training in ergonomic aspects in both LAP and RAS [43].

Joint analysis of spectrum and analysis graphs has been widely applied in scientific literature for the analysis of muscle fatigue [44]. This method relates muscle activity and the mean frequency of the EMG spectrum, obtaining a schematic representation of the relationship between localized muscle fatigue and force [45]. Previous studies have studied trapezius muscle fatigue during LAP, as the trapezius is frequently vulnerable to localized muscle fatigue [46]. This fact is being corrected with RAS, as the activity of the main dominant muscles in LAP has been reduced [47]. Furthermore, these results showed less localized muscle fatigue and greater use of force in RAS than in LAP in novice surgeons. However, in the case of expert surgeons, the use of force and localized muscle fatigue was lower in LAP. This fact could be related to their previous experience in LAP [43]. However, it would be necessary to take into account other aspects that may affect surgeons’ ergonomics, such as the ergonomically optimal configuration of the working environment, their prior training in the field of surgical ergonomics, activities performed prior to the study, or the use of simulators of similar surgical environments, among others.

In general, all muscles bilaterally show statistically significant differences between the LAP technique and RAS. This fact demonstrates the different use of muscles between LAP and RAS for both types of surgeons, experts and novices [39]. Similarly, obtaining better results by applying MLR as an AI technique could be related to the limitations of using MLP, such as its black box nature or the need to use a large amount of data for training.

Considering more specifically the results for the tasks performed by the LAP technique (Figure 6), for the dissection task, the left brachioradialis showed lower muscle activity for novice surgeons, and the right medial gastrocnemius showed lower muscle activity for expert surgeons. For the peg transfer task, the muscles that did not follow the general trend were the medial gastrocnemius bilaterally, showing less muscle activity for novice surgeons, and the right vastus lateralis, also showing less muscle activity for novice surgeons. In the same way, the results obtained for the suturing task followed the general trend, except for the medial gastrocnemius bilaterally, displaying less muscle activity for novice surgeons, highlighting the similarity for the middle trapezius bilaterally.

In the case of the tasks using RAS (Figure 6), the dissection and labyrinth tasks followed the general trend, with less muscle activity being shown by novice surgeons, except for the right erector spinae, which showed less muscle activity for the expert surgeons. For the peg transfer task, the general trend was followed, with less muscle activity for novice surgeons, except for the erector spinae bilaterally and right middle trapezius, which displayed less muscle activity for expert surgeons. Finally, for the suturing task, the general trend was followed, with less muscle activity for novice surgeons, except for the erector spinae bilaterally and the right middle trapezius and right upper trapezius, which displayed less muscle activity for expert surgeons.

It is noteworthy that, in general, muscle activity on the right side is lower than on the left side, both for experienced and novice surgeons. This fact could be related to the dominant side of surgeons, who require less effort to complete a manual task [48]. Consistent with this fact, muscle activity on the dominant side also appears to increase slightly over time compared to the nondominant side, suggesting that the dominant side may be more susceptible to developing muscle fatigue and possible future musculoskeletal disorders [49]. In the same way, the results obtained for Figure 7 are consistent with those obtained in previous analyses [50].

The results from Figure 8 could be related to the variability of the surgeon’s surgical experience and the stress generated during the performance of surgical activities [51]. Moreover, in almost all cases, a high to excellent relationship was achieved (R²>0.75), highlighting that for the dissection task, expert surgeons with RAS achieved the highest R² values for most parameters for the labyrinth task, expert surgeons with LAP, for the peg transfer task, novice surgeons with LAP and for the suturing task, expert surgeons with RAS. These positive predictive results are in agreement with previous studies that also showed high to excellent predictive results performing LAP and RAS [5]. As for the previous results, the predictive results for Figure 9 are in agreement with previous studies [5].

As in previous sections, these satisfactory predictive results for Figure 10 are in agreement with previous studies [5]. Moreover, in almost all cases, a good to excellent relationship was achieved (R²>0.5), highlighting that for the dissection task, expert surgeons with RAS obtained the highest R² values for most parameters, for the labyrinth and pin transfer tasks, novice surgeons with LAP, and for the suturing task, expert surgeons with RAS. In general, the outcomes of this study exhibit a slight superiority over those obtained in preceding studies that used AI models for localized muscle fatigue prediction [52-55].

Among the limitations of the present study is the small number of participating surgeons. Although previous studies have established the relationship between muscle activity and localized muscle fatigue experienced during simulated tasks compared to surgical procedures [19], further research in this area is warranted. This should involve increasing the number of surgeons involved and conducting more intricate surgical procedures that closely resemble real-world scenarios, obtaining more definitive and representative results [56]. On the other hand, the limitations due to the number of EMG sensors available and that can be used simultaneously have prevented us from recording the muscle activity of all the muscle groups of interest. For future work, we intend to include recordings of other muscle groups, such as the deltoid or biceps brachii, which have also been shown to be related to LAP activity. Furthermore, the incorporation of additional wearable and physiological sensors, such as EDA or an electrocardiogram, should be considered in subsequent studies to enhance the representativeness of the results.

For future research, the predictive models derived from this study are intended to be integrated and validated within surgical training activities and in actual surgical practice across diverse specialties, including general surgery, urology, and gynecology, among others. This integration will contribute to the prevention of musculoskeletal issues during surgical practice and, consequently, enhance the quality of surgical outcomes. Furthermore, it would be beneficial for future studies to include participating surgeons with prior experience in RAS and to analyze the impact of experience on the ergonomics of surgeons who perform robotic surgery. Additionally, it would be insightful to investigate the correlation between the onset of localized muscle fatigue and the quality of surgical performance during RAS surgery.

Similarly, the development of real-time feedback systems would allow surgeons to be alerted to situations of high localized muscle fatigue and prevent possible musculoskeletal risks during surgical practice. In this way, predictive models could be integrated into surgical training and actual clinical practice, improving the health of the surgical team and, consequently, the quality of patient care.

In this study, a novel predictive model based on MLR and MLP has been developed to predict the localized muscle fatigue of a set of surgeons’ muscle groups during LAP and RAS. This model achieved satisfactory R² and RMSE coefficients for all parameters extracted from the EMG signal, predicting with high accuracy the localized muscle fatigue values. The MLR predictive model demonstrated superior performance relative to the MLP model. Therefore, it is a promising starting point for predicting surgeons’ ergonomic risks during surgical practice in MIS.

Regarding the analysis of localized muscle fatigue, in general, RAS showed less muscle fatigue for novice surgeons compared to LAP practice, although it was higher for expert surgeons. Regarding ergonomic aspects, postural habits and motor control are different among surgeons with different surgical experience, which affects muscle activity patterns. The prevention and improvement of ergonomics in surgical practice minimizes localized muscle fatigue and forced postures, improves the health of surgeons, favors surgical performance, and consequently, contributes to improving the quality of patient care.

Ergonomic analysis makes it possible to study individual variations in the different ergonomic parameters according to the surgeons’ experience or the surgical technique used, thus allowing for improved training and surgical guidelines. Considering the rapid expansion of RAS in clinical practice, the analysis and improvement of the ergonomic conditions of surgeons and the surgical team are essential to provide optimal working conditions, ensuring the well-being of the surgical team and the quality of patient care. The training of the surgical team in the optimization of ergonomic conditions is fundamental to maximizing the benefits of MIS for both patients and surgeons.